Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,83 @@
+---
+license: mit
+pretty_name: BoLT Parallelism Configuration (16 GPUs)
+task_categories:
+- other
+tags:
+- bayesian-optimization
+- benchmark
+- parallelism-configuration
+- tabular
+size_categories:
+- n<1K
+---
+
+# pco16
+
+Throughput of distributed LLM training as a function of the parallelism
+configuration, on 16 GPUs across 2 hosts.
+Tabular benchmark for the `bolt` problem `pco16`: the 919
+rows are the full candidate set.
+
+- **Objective** `throughput_mean`, to maximise. `NaN` where the run OOMed, since
+  no throughput is observed at all -- a hidden (crash) constraint, not a bad value.
+- **Constraint** `ran_successfully`. 632 of 919
+  configurations run; the rest exhaust GPU memory. Feasibility is only learnable
+  by attempting the run, so it cannot be decoupled from the objective.
+- **Auxiliary** `peak_mem` (GB, `inf` when infeasible), `peak_mem_normalised`
+  (the run's own feasibility margin, `< 0` iff feasible, a fixed pad value when
+  infeasible) and `overall_time` (wall clock in seconds, `NaN` when infeasible).
+
+`peak_mem` does **not** define the constraint: feasible runs here reach
+70.6 GB, so no memory threshold separates the
+classes. `peak_mem_normalised` does, by construction.
+
+## Instance context
+
+Constant for every row; these describe the instance, not the search space.
+
+| field | value |
+|---|---|
+| `num_layers` | 32 |
+| `num_attention_heads` | 32 |
+| `hidden_size` | 4096 |
+| `ffn_hidden_size` | 11008 |
+| `batch_size` | 512 |
+| `seq_length` | 2048 |
+| `log2_num_gpus` | 4 |
+| `log2_num_hosts` | 1 |
+
+## Columns
+
+The table is the raw sweep export, unchanged. Every knob halves or doubles, so
+the sizes arrive as `log2_*` and their levels are evenly spaced. `zero_stage` is
+the raw stage (0/2/3); `bolt` searches it as an ordinal 0/1/2.
+
+| column | role | meaning |
+|---|---|---|
+| `log2_dp_size` | input | log2 of data-parallel size |
+| `log2_tp_size` | input | log2 of tensor-parallel size |
+| `log2_pp_size` | input | log2 of pipeline-parallel size |
+| `log2_cp_size` | input | log2 of context-parallel size |
+| `log2_sp_size` | input | log2 of sequence-parallel size |
+| `log2_dp_bucket_size_mb` | input | log2 of DDP gradient bucket size in MB |
+| `zero_stage` | input | ZeRO stage, one of [0, 2, 3] |
+| `log2_num_model_chunks` | input | log2 of number of model chunks |
+| `log2_grad_accum_steps` | input | log2 of gradient accumulation steps |
+| `ran_successfully` | output | ground-truth constraint label |
+| `throughput_mean` | output | objective, `NaN` when infeasible |
+| `throughput_std` | output | spread over the timed steps, `NaN` when infeasible |
+| `peak_mem_normalised` | output | feasibility margin, `< 0` iff feasible |
+| `overall_time` | output | wall clock in seconds, `NaN` when infeasible |
+| `all_time_steps` | output | per-step times, as a string list |
+| `peak_mem` | output | peak GPU memory in GB, `inf` when infeasible |
+| `peak_mem_per_gpu` | output | per-GPU peak memory, as a string list |
+
+## Best observed
+
+`throughput_mean` = 0.14894 at `log2_dp_size`=4, `log2_tp_size`=0, `log2_pp_size`=0, `log2_cp_size`=0, `log2_sp_size`=0, `log2_dp_bucket_size_mb`=8, `zero_stage`=2, `log2_num_model_chunks`=0, `log2_grad_accum_steps`=2
+(peak memory 57.5 GB).
+
+## Provenance
+
+`data_32depth-16gpus.csv`, converted to parquet unchanged.